Cloud & InfrastructureFreeactiveReviewed by MCPIndex

AWS MCP Server

AWS agent tooling under IAM — service discovery, CloudWatch analysis, documentation, and account operations require explicit least privilege.

Looking for more MCP servers? Browse the full MCP tools directory or explore more tools in Cloud & Infrastructure.

Reviewed by MCPIndex

MCPIndex assessment

AWS MCP is an IAM architecture exercise disguised as an integration. Its value is exceptional for cloud incident triage, but its safety is exactly the policy attached to the principal — no prompt can compensate for AdministratorAccess. Deploy one read-only role per account, force identity and region verification, and audit every call through CloudTrail. Anything else is a production experiment.

Quick overview

AWS MCP is not an AWS assistant; it is an IAM principal with a language model attached. That framing is the only safe starting point. The server can give an agent structured access to AWS services, CloudWatch telemetry, documentation, and account resources, compressing the usual console-and-CLI incident workflow into a cross-service reasoning pass. An agent can correlate an ECS deployment with an ALB target-health failure, inspect the relevant CloudWatch logs, and identify an IAM or networking mismatch without you pasting five screens of output. The architectural power comes from AWS's own control plane: every action is evaluated by IAM, CloudTrail records it, and service-specific policies can restrict the surface. The danger is equally native: a broad AdministratorAccess policy turns a helpful diagnosis into a natural-language root account, while a restrictive policy often produces AccessDenied responses that the model misinterprets as resource absence. The professional deployment pattern is a dedicated role with read-only, resource-scoped permissions, an explicit account and region preamble, and a separate human-approved write path. If the agent cannot tell you which account, region, principal, and policy allowed a result, it is not ready for production.

Best for

The teams and workflows that benefit most from this tool.

CloudWatch-grounded incident diagnosis

Read-only AWS inventory and configuration audits

Cross-service troubleshooting across ECS, Lambda, EC2, and networking

AWS documentation and account-aware operational assistance

What this MCP server is best for

  • Inspecting AWS resources, environments, and service configurations through AWS MCP Server.
  • Reviewing EC2 instances, S3 buckets, IAM scoping, and cloud setup details during debugging workflows.
  • Helping cloud and DevOps teams reduce context switching when managing AWS infrastructure.

When to choose it

Choose AWS MCP Server when you want an MCP server focused on Cloud & Infrastructure and need tighter integration with your existing tools.

Good fit

awscloudiamcloudwatchec2ecs

Limitations

Things to watch before choosing this tool.

Broad IAM permissions create catastrophic blast radius

Account and region drift returns plausible empty results

CloudWatch queries can be costly and context-heavy

Multi-account deployments require explicit role and trust management

AWS MCP Server Configuration

Use the following configuration as a starting point for Claude Desktop or any compatible MCP client, then replace placeholder credentials with your own values.

claude_desktop_config.json
{
  "mcpServers": {
    "aws": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "AWS_PROFILE", "-e", "AWS_REGION", "public.ecr.aws/aws-mcp/aws-mcp-server:latest"],
      "env": {
        "AWS_PROFILE": "mcp-readonly",
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

How to set up AWS MCP Server

These setup steps cover the typical installation flow for this MCP server.

  1. 1

    Create a dedicated IAM role or profile with read-only, resource-scoped permissions for the services in scope — start with CloudWatchReadOnlyAccess and service-specific read policies, not AdministratorAccess.

  2. 2

    Use a separate AWS profile named mcp-readonly and pass AWS_PROFILE and AWS_REGION explicitly to the server process.

  3. 3

    Pro-Tip: The fatal flaw 90% of developers commit is letting the agent inherit the default AWS profile and region. The credentials are valid, so the agent can inspect or mutate whichever account you touched last; the region mismatch can return empty resource lists without an error. Require the first prompt to print account ID, caller ARN, and region, and deny all write actions in IAM until a human approval path exists.

  4. 4

    Enable CloudTrail monitoring for the MCP role and create a budget or anomaly alert for unexpected API usage — agent loops can generate many Describe and List calls even when nothing is wrong.

  5. 5

    For production diagnosis, use a read-only role in each target account and assume it through a tightly scoped trust policy; never distribute long-lived access keys to local MCP configs.

  6. 6

    Verify with: identify the caller, list the account and region, then inspect one known resource. A correct setup proves identity before returning infrastructure facts.

Compatibility

Supported environments for AWS MCP Server.

Claude Desktop

Yes

Cursor

Yes

VS Code

Yes

Run with a dedicated read-only AWS profile or role. Always verify caller identity, account ID, and region before resource queries.

Frequently asked questions

Common questions for AWS MCP Server.

The agent reports that a resource does not exist, but the AWS console shows it clearly — no error appeared. Why?

The silent failure is usually account or region drift, followed by IAM filtering. AWS APIs return an empty list when you query a valid region where the resource is absent; an insufficient policy may also deny a specific discovery call while the wrapper reduces it to no results. The engineering fix is to require sts:GetCallerIdentity, account ID, region, and caller ARN at session start, then confirm the target service's region and IAM permissions before treating an empty result as proof of absence.

Can AWS MCP deploy or delete infrastructure?

Only if the attached IAM policy allows it. Do not rely on prompt instructions. Remove write, delete, and pass-role permissions from the agent role and route mutations through reviewed IaC pull requests or an explicit human approval service.

Why are CloudWatch results expensive or slow?

Broad log and metric queries fan out across time ranges and streams. Bound the interval, filter by service and severity, cap returned events, and query metrics before logs to narrow the search space.

Does it work across multiple AWS accounts?

Yes through role assumption, but each account and role is a separate trust decision. Expose one clearly named profile or endpoint per account so the agent cannot confuse staging and production.

AWS MCP Server vs Competitors

FeatureAWS MCP ServerCompetitor
Cross-service reasoning Correlates IAM, CloudWatch, compute, and networking context Single-service dashboards with isolated context
Authorization Native IAM policy evaluation and CloudTrail audit Connector-specific scopes with uneven coverage
Account safety Requires explicit account and region grounding Single-project SaaS connectors reduce account ambiguity
Infrastructure mutation Possible if IAM permits it Read-focused observability integrations block writes by design
Cloud breadth AWS-native services and APIs Multi-cloud abstraction across providers

Related Guides